Potentiale von Machine Learning Modellen zur Prognose von Lastgängen bei Fertigungsprozessen

Ellerich, Max; Schmitt, Robert H. (Thesis advisor); Kampker, Achim (Thesis advisor)

1. Auflage. - Aachen : Apprimus (2021)
Book, Dissertation / PhD Thesis

In: Ergebnisse aus der Produktionstechnik 4/2021
Page(s)/Article-Nr.: xviii, 213 Seiten : Illustrationen, Diagramme

Abstract

The climate crisis requires a global rethink of our use of resources and, in particular, a reduction in global CO2 emissions. The policy of energiewende is activating a fundamental transition to completely CO2-neutral energy production in the mid-term. This transformation, however, will ultimately lead to increased volatility of the power grids. By some estimates, manufacturing and production companies account for a substaintial 47 % of the demand for electrical power in Germany. Therefore, the focus of this paper is on the consumer-side adjustment of the electricity demand of companies in this sector to the incipient volatility of electrical power grids. To adjust their electricity demand to available power, companies must know the load curves of individual production steps prior to production. Only then companies may effectively plan their orders a priori so the addition of the required power of processing stations working in parallel corresponds to the available electrical power. In order to develop a solution that can be transferred to numerous processing stations in a scalable way, this thesis shows an approach that allows the performance prediction of processing steps with the help of a data-based black-box model. This approach predicts both the total energy of a machining step and the corresponding load curve. First, existing models and approaches are compared to derive the research gap. Based on the requirements of energy-sensitive production planning, the model itself is derived and its application in the business context is described. The model is validated using two case studies from the fields of 3D printing and plant engineering. Since the load curve of all processing technologies can be transferred to the selected approximation method Energy Blocks, the approach has the potential to predict load curves for a wide array of technologies and serves as a first step for energy-flexible production planning.

Identifier